The Personal Website of Mark W. Dawson
Containing His
Articles, Observations, Thoughts, Meanderings,
and some would say Wisdom (and some would say not).
Is Artificial Intelligence Possible?
Introduction
Curiosity, inquisitiveness, interest, questioning, querying, searching, creativity, insights, and inquiry are all part of being intelligent. The search for answers to the questions of good from evil, right from wrong, truth from falsehood, creative from destructive, reasonable from emotional, love from hate, wisdom from folly, and beauty from ugliness are part of being intelligent. The search for the knowledge of who, what, when, where, why, and how of our universe are also part of being intelligent. All these items are core questions that intelligence has been searching for since the dawn of mankind. Can Artificial Intelligence even understand and ask these questions, let alone find answers? Could Artificial Intelligence ever create beautiful works of paintings, sculpture, architecture, music, literature, or poetry? Can Artificial Intelligence make startling new discoveries in science and technology, or just make innovative improvements in current science and technology?
We must differentiate between Artificial Intelligence and Automated Reasoning. In computer science, artificial intelligence (AI), sometimes called machine intelligence, is intelligence demonstrated by machines, in contrast to the natural intelligence displayed by humans. Leading AI textbooks define the field as the study of "intelligent agents": any device that perceives its environment and takes actions that maximize its chance of successfully achieving its goals. Colloquially, the term "artificial intelligence" is often used to describe machines (or computers) that mimic "cognitive" functions that humans associate with the human mind, such as "learning" and "problem-solving". Automated reasoning is an area of cognitive science (involving knowledge representation and reasoning) and metalogic dedicated to understanding different aspects of reasoning. The study of automated reasoning helps produce computer programs that allow computers to reason completely, or nearly completely, and automatically. Although automated reasoning is considered a sub-field of artificial intelligence, it also has connections with theoretical computer science and even philosophy.
Many people conflate the terms of Artificial Intelligence and Automated Reasoning. They are not equivalent, as Artificial Intelligence does much more than Automated Reasoning, and Automated Reasoning is much more advanced than Artificial Intelligence. Computerized medical diagnostics is one of the most advanced capabilities of Automated Reasoning. To make the leap to Artificial Intelligence in medicine then the Artificial Intelligence would have to start when a patient arrives for a possible medical problem until they are discharged after the treatment of their medical problem, with the Artificial Intelligence directing all the activities regarding the patient care. There are many areas in this process where Automated Reasoning is applied with success, but Artificial Intelligence is the start to end control of this process through non-human (Artificial) Intelligence.
Automated Reasoning utilizes both Formal and Informal Logic to reach its conclusions. But even using both formal and informal logic can lead to an invalid conclusion by the introduction of Formal Fallacies or Informal Fallacies. Automated Reasoning also uses complex Boolean algebra to reach its conclusions, and if this algebra is incorrect than the conclusion is incorrect. Therefore, if an Automated Reasoning reaches a conclusion that contains any logical fallacies or incorrect algebra than the conclusion will be wrong. It is, therefore, very important that you be wary of any Automated Reasoning conclusions until you can be assured that the reasoning contains no errors.
Artificial Intelligence implies that no human intervention is required to produce human-like intelligent capabilities from Artificial Intelligence. It is this broad definition of Artificial Intelligence that I wish to examine in this article. This short article attempts to explore the boundaries of Artificial Intelligence and provide insights into the answers to the question of “Is Artificial Intelligence Possible?”. This will be done by providing several examples of the nuisances of the problems that highlight the issues of Artificial Intelligence.
Computing Complexities
The Traveling Salesman, Hamiltonian Cycle, Set Partitions, Subset Sum, and Satisfiability problems in Computer Science, which are all interconnected, are difficult issues that have not been resolved. However, for true Artificial Intelligence, these problems must be resolved for Artificial Intelligence to succeed. A brief look at two of these problems is illuminative.
The Traveling Salesperson Problem
The Traveling Salesperson problem is fairly simple to explain but very difficult for a computer to solve. A traveling salesperson is tasked with visiting five clients across the continental United States then returning home. He must schedule the route to travel the shortest distance, take the least amount of travel time, and have minimal transportation costs. You must also factor in the mode of transportation i.e. (airplane, railroad, bus, or automobile) on this sales trip. In this case, there are four criteria (distance, time, cost, and transportation mode) that need to be resolved. Mathematically there are 720 possible routes that can be taken, and it would be easy to determine the shortest distance amongst all these routes. But for each of the 720 routes you then need to factor in the travel time, which is dependent on the mode of transportation, then factor in the costs for each possible route which is also dependent on the mode of transportation. You must do this for all possible combinations of distance, time, cost, and transportation mode to arrive at the best schedule for the visits.
For a digital computer to accomplish this feat it may take several days or weeks to compute the best schedule. Whereas the newer Quantum computers may be able to accomplish this feat in several hours or perhaps a day or two. However, if you generalize this problem by not restricting it to five visits, but by any possible number of visits, the more visits to be calculated the more the computational time increases in a logarithmically and not linearly manner. The computational time required for each additional visit takes more than twice the amount of time to calculate than the previous number of visits. Very quickly the computation starts to take years, decades, or centuries to accomplish when you have hundreds, thousands or tens of thousands of routes or additional factors.
Yet the human mind can solve this problem in a relativity short amount of time for a limited number of visits. This is because human intelligence can grasp which routes can be rejected as unlikely to be satisfactory, and which routes look promising. The human mind then concentrates on the few routes it has selected as promising, then applies its knowledge and experience of the travel factors to pick the best route possible. Mathematically, the route the human mind has chosen may not be the best possible, but the human mind makes a value judgment if it is worth spending additional time to determine the best possible route. The computer must compute all the possibilities to determine the route that should be taken. This could be a time consuming and expensive process to accomplish and may not be worth the effort. However, many package delivery companies are attempting to accomplish this with Artificial Intelligence. They succeed, somewhat, by placing artificial constructs to limit the possibilities, which reduces the time and complexity to determine the route. Even then, the driver must make adjustments during the delivery route to compensate for unknown factors (road closings, road construction causing delays, accidents, traffic jams, choke points, etc.), as well as their intuitive knowledge of better alternatives.
How exactly the human mind does this is unknown, and it is a field of study by both computer scientists and human intelligence scientists. The question, as regarding this article, is it possible for Artificial Intelligence to have this capability to selectively choose or reject possibilities for further examination? And can Artificial Intelligence have foreknowledge of unknown factors to adjust the route? The solution of the Traveling Salesperson problem is required for true Artificial Intelligence to be possible.
The Satisfiability Problem
The Traveling Salesman Problem is actually a subset of the Satisfiability Problem, which is a very large problem in Computer Science. Much time, effort, resources, and monies are spent trying to solve the Satisfiability Problem. In essence, the Satisfiability Problem is where a number of criteria need to be resolved and correlated to reach a satisfactory conclusion. The resultant of each individual criterion may be difficult or time-consuming to reach, but then you need to compare the results of each criterion with all the other criteria, in all possible combinations, to achieve a satisfactory answer.
The example of scheduling NFL season games is illustrative of this problem. Two white papers that examine this issue are “National Football League Season Scheduling” and “Prime Time: The NFL optimizes its playing schedule”. Even with the advanced technologies that these white papers describe the computer does not reach a conclusion. Several hundred possible solutions are identified, which are then reviewed by the NFL scheduling committee composed of knowledgeable and experienced persons to reach a final schedule. This is an example of Artificial Intelligence being helpful but not definitive. Artificial Intelligence is utilized to obtain several hundred possible solutions, but humans are required to decide which of these several hundred solutions is the best schedule. Even with all this effort, there are still many complaints about the NFL schedule, which probably points to the human side of the issue, but maybe because of the lack of enough criteria or specificity of the criteria in the computer automated process. In either case, computer science cannot solve the Satisfiability Problem, it can only assist in solving the problem.
And the solution to the Satisfiability Problem is something that is needed to achieve true Artificial Intelligence.
Understandability Problems
Understandability is what makes us intelligent. We remember the past, perceive the present in many different perspectives, and adjust our future words and deeds based on the past and present. Artificial Intelligence, therefore, must be able to do this to be truly intelligent. Yet there are many aspects to understandability that pose difficulties for Artificial Intelligence. Here are just a few of many understandability difficulties regarding Artificial Intelligence:
The Comprehension Problem
Does Artificial Intelligence comprehend the meaning of what is occurring? How often have we seen a photograph or video of a soldier returning from active duty, being greeted by their child, in the presence of their spouse? Artificial Intelligence can analyze this photo/video and determine many things about the photo/video. It can identify the people or animals in the photo/video, perhaps the people’s sex and age group depending on the photo/video angle, aspects of their people's clothing, perhaps their surroundings, and other objects in the photo/ video. But Artificial Intelligence can make mistakes, often with hilarious consequences like confusing the sex of a person, or mislabeling a person, but also with the potential of tragic consequences like a self-driving car that makes a mistake that results in an accident that could cause serious bodily harm or even death.
But besides this issue can Artificial Intelligence comprehend the meaning of the photo/video? A Human Intelligence would realize the above-mentioned photo/video is a homecoming of a soldier parent that has been separated from their family by going into harm’s way. The possibility that the returning soldier may never have returned, or returned with a serious injury, makes the homecoming joyous. A human would expect that this family will then retrieve the baggage of the soldier, and head home for a celebration of the return of the soldier. Can Artificial Intelligence determine all of this? If this photo/video had occurred on a basketball court, and the family was surrounded by the players/coaches, would Artificial Intelligence make a determination of a celebration of something that had occurred in the basketball game? Determining the context of the events that are occurring, relating them to past events, and projecting future events based on the present and the past events is important to comprehension.
Speech recognition has many difficulties as well. Given these difficulties, I am amazed that it works as well as it does. Speech has many subtleties associated with understanding. Short and concise speech is often properly recognized, but longer and more complex speech can be difficult for Artificial Intelligence to understand. This is because language itself has many components that are subject to experience and interpretation. Many times, when we speak, we leave information out that we know the listener can fill in based on their human experience and the context of what we are saying. Because of this, it is not necessary for us to be precise in all we say and still be understandable. We will often use an incorrect word that would give an imprecise meaning to what we say, with the understanding that the listener will properly interpret the incorrect word. On occasion, we will make up a word that expresses what we are thinking, and the listener will understand our meaning.
One of the other properties of speaking is nonverbal communication. A look or expression when you are saying something gives context to what you are saying. For instance, I could be deliberately saying something outrageous and the expression on my face reveals that I am being facetious or sarcastic, thus negating what I say. The old joke of Latin people speaking with their hands bears some truth. Our physical exertions when speaking add context and emotional meaning to what we are saying. Even our body position can add context. Speech recognition cannot determine any of this and thus lead to a wrong interpretation of what we are saying.
Humans can comprehend what we are saying despite the above issues. Can Artificial Intelligence comprehend what is being said under the same circumstances? Today, when we utilize speech recognition it is in a verbal only manner. We say something into a microphone and the speech recognition interprets what we say and takes some action. We have trained ourselves to speak clearly and concisely in not-ambiguous words so that the speech recognition works properly. But this is not true speech recognition, but a human workaround to compensate for the limitations of Artificial Intelligence speech recognition.
Can Artificial Intelligence perceive the meaning of the present, glean the past from the present, and project the possible future from the past and present? Can Artificial Intelligence interpret the totality of a situation to determine what is truly happening? These are all aspects of comprehension and thus intelligence, and until Artificial Intelligence can do this it is not truly intelligent.
The Insights Problem
As seen by the Traveling Salesperson problem both insight and selective reasoning is important for intelligence – both human and Artificial Intelligence. The following true story highlights these factors.
Galileo Galilei as a young man had not yet established himself as a scientist. Indeed, he was trained in the medical sciences of his time, but his curiosity went beyond medicine. One day he attended church services at the Pisa Cathedral and became bored by the service. He gazed at his surroundings for something of interest. The Cathedral was packed with other parishioners who were surrounded by the sights of the Cathedral.
Biblical scene paintings and sculptures, as well as Sculptures of Christ, the Apostles, and the Saints, and the lamps to illuminate all of this were all around him. Priests were vocally conducting the services along with a choir. Stain glass windows illuminated the Cathedral showing some rainbow effects. Candles were lit and lamps were lighted throughout the Cathedral. The air currents within the Cathedral spread the aroma of the people and candles and lamps, as well as the swaying of the lamps that hung from the ceiling. The ceiling itself was an architectural marvel. Galileo had a rosary in one hand and a cross in the other. Communion was given so that he had the taste of the wafer. In essence, all his senses were active as he gazed around him. In the midst of all this Galileo made a discovery that astonished the scientific world and established himself as a leading scientist.
What was it that Galileo had discovered? Before I answer that question, I would like you to imagine that standing next to Galileo was a human appearing Artificially Intelligent android that was receiving the same sensory information as Galileo. Also assume that this android had no more, nor less, scientific knowledge than Galileo. The question then becomes “What was it that Galileo had discovered, and did the android discover it as well?”. This is a key question in Human Intelligence and Artificial Intelligence.
Galileo focused in on the swaying of the lamps suspended from the ceiling. He noticed that the time it took for the swing of the arc of the lamp seemed to be the same no matter the length of the arc. He timed these swings by using his resting pulse, as clock and watches were not yet invented and could not be invented until Galileo made his discovery. Longer arcs took the same amount of time as smaller arcs. This seemed strange to him and he began a scientific investigation to figure it out. He discovered that it did not matter what the length of the arc was, nor the weight of the bob, made any difference in the swing time. The only variable was the length of the rod, as shorter rods took less time than longer rods. He carefully measured this effect and created a mathematical formula to calculate the movement of pendulums. This discovery was a scientific breakthrough in many different ways. It helped establish the scientific method of observation, experimentation, and measurement, and the utilization of mathematics to describe the results. It also led to the invention of clocks that could measure smaller increments of time to greater accuracy. This not only impacted science but also all human activities that need clocks of greater precision and accuracy (which is most of human activities).
Which brings us back to the question of would the “Artificially Intelligent android discover it as well?”. No one knows if this is possible, as the human ability to filter out and focus on something of interest, then have an insight into the interesting something, is not understood. Yet for progress to occur in an intelligent society you need to have this ability, so Artificial Intelligence must have this ability.
The Creativity Problem
One of the most impressive sculptures in the world is “David” by Michelangelo. To behold this sculpture is to be moved by the beauty of its magnificence. When one of Michelangelo’s patrons asked him how he could create such beauty from a slab of marble he responded that he just saw David in the marble and removed the parts that weren’t David.
If an Artificial Intelligence looks at a slab of marble will it see anything other than a slab of marble? Will Artificial Intelligence do anything with a slab of marble that is creative and not done before? Will Artificial Intelligence ever be able to create music such as Beethoven or Stravinsky, paintings such as DaVinci or Picasso, literature such as Shakespeare or Tolstoy, or any other magnificence work of art and artists? Would Artificial Intelligence produce insights such as Newton or Einstein, Euler or Gödel, Mendel or Crick and Watson, Descartes or Sartre, to name but a few of the intellectual genius’s that have walked the earth?
To create something from nothing, to think what no others have not thought before, and myriad other human creative activities are intelligence at its finest. Can Artificial Intelligence accomplish this? To examine this issue, we must examine the issue of the Brain versus Mind.
The Brain versus the Mind
The Brain and the Mind are two different aspects of intelligence. Your brain is part of the visible, tangible world of the body. The Brain is the structure, interconnections, and the processing and storage of information from our senses and body. Your mind is part of the invisible, transcendent world of thought, feeling, attitude, belief, and imagination. The brain is the physical organ most associated with the mind, but the mind is not confined to the brain. The Philosophy of Mind is the branch of philosophy that studies the nature of the mind, mental events, mental functions, mental properties, consciousness and their relationship to the physical body. The mind-body problem, i.e. the relationship of the mind to the body, is commonly seen as the central issue in philosophy of mind, although there are other issues concerning the nature of the mind that do not involve its relation to the physical body.
We are gaining much more knowledge about the physiology of the brain, but we are still groping to gain knowledge of how the mind works. We have only one model of how an intelligent mind works – the human mind. To construct an Artificial Intelligence with a mind of its own requires that we understand how the human mind works. But until we can understand how the human mind works it may not be possible for us to invent true artificial intelligence. And Artificial Intelligence without a mind only allows you to work within known parameters. It cannot go beyond the known into the unknown and comprehend, and to be insightful and creative without a mind.
Is learning a function of the mind or the brain, or perhaps some combination of mind/brain? Many animals learn basic functions that are required for survival. But they go no further than what is required for survival. Human intelligence learns so much more than how to survive. It learns many things and expands its knowledge beyond what is required for survival. Currently, much of Artificial Intelligence learning is at the direction of humans (just as adult humans teach children) and is limited in scope to the task at hand. Human learning is about learning many different things, how to think about what they have learned, and how to apply what is learned to their words and deeds. Eventually, a child begins to learn on their own and of their own volition. For Artificial Intelligence to be possible it must be able to do all of these things about learning and at its own volition.
Artificial Intelligence has learned to be very good at specific tasks. But this learning has not been able to be transferred to other tasks. Although the techniques that they have utilized to achieve these goals can be applied to other goals, what they have learned when doing these tasks cannot be transferred to other endeavors. For instance, Artificial Intelligence's ability to play Chess or Go has made tremendous strides to the point that Artificial Intelligence now regularly beating human opponents. However, what they have learned from playing Chess cannot be applied to playing Go, and vice-versa, let alone to other tasks outside of Chess and Go. Humans learn how to think, then apply this learning on how to think to other tasks, and these other tasks may not be related to each other. Until Artificial Intelligence can learn unknown things on its own volition, and transfer this learning to other endeavors, it cannot be truly intelligent.
This leads us to the questions of how the human mind/brain thinks and learns? Of this we are uncertain, and this uncertainty applies to how we can accomplish this with Artificial Intelligence. These questions about how the human mind works need to be discovered to be applied to Artificial Intelligence. There is also the possibility that our attempts to duplicate this in Artificial Intelligence may help us to better understand how the human mind works. However, until we achieve this understanding of the human mind it may not be possible to create Artificial Intelligence.
Morality and Ethics
This section is not so much about the possibility of Artificial Intelligence, but the desirability of Artificial Intelligence. The questions of morality and ethics have bedeviled mankind since they have learned to think. There are no easy answers to these questions, and there are many dilemmas, quandaries, conflicts, and predicaments that lead to unsatisfactory answers.
One of the most famous and easily understood questions is the runaway trolley problem. Assume that a trolley has run away and has lost its braking ability. Down the track are five workmen with their backs turned to the trolley who will be run over and killed it nothing is done. The trolley engineer notices a sidetrack that they can divert to, but this sidetrack has one person standing with his back turned. Do you divert to the sidetrack and kill one person, or stay on the main track and kill five people? Most people would say you should divert until you tell them the workman on the sidetrack is a family member. This dilemma can be rephrased in many ways and less personal ways, but the dilemma occurs in one form or another that does not make for an easy answer.
So much of human philosophy, theology, morality, and ethics questions deal with these problems, with many of the answers unresolved. Yet if we do create Artificial Intelligence it will encounter these issues while processing its tasks. How will Artificial Intelligence resolve these issues, especially if we cannot provide guidance to Artificial Intelligence during its invention? Can we create guidelines for Artificial Intelligence morality and ethics, and will these guidelines be effective?
The example of HAL in the movie “2001: A Space Odyssey” is illuminative. HAL, the AI computer, was programmed to be honest with the astronauts at all times. However, HAL was also programmed to not tell the astronauts the true purpose of their mission. This set-up a dilemma that drove HAL to kill all but one of the astronauts. The famous robot stories of Isaac Asimov have different dilemmas. The Three Laws of Robotics were created for robots to protect and serve humans, at the same time protecting the robot from harm or destruction. The many short stories and a few novels dealt with the dilemmas of these laws when the robot was performing their tasks. In many cases, it did not work out well for humans or the robots in these stories.
There are many other issues and concerns with the introduction of Artificial Intelligence in human society (see the “Top 9 ethical issues in artificial intelligence” for some examples). Too many issues and concerns to outline here. However, these issues and concerns need to be addressed before we turn over most of society to Artificial Intelligence. You should also remember my cautions of “Change and/or New“ and the “Law of Unintended Consequences” and its outcomes of unexpected benefits, unexpected drawbacks, and perverse results.
Conclusion
The development of Artificial Intelligence has provided great benefits to mankind. Problems and the solutions to these problems have been obtained by the utilization of Artificial Intelligence. But is Artificial Intelligence really Intelligence, or simply an extension of human intelligence that allows for the storage of vast quantity of facts, the interconnections of these facts, and the fast retrieval and logical processing of these facts, i.e. Automated Reasoning. The results of this Automated Reasoning are then utilized by Human Intelligence to expand human capabilities.
A very good book on Artificial Intelligence, that is readable and understandable by the general public, and worth the read is “Artificial Intelligence: A Guide for Thinking Humans” – by Melanie Mitchell. An entire book “ The Outer Limits of Reason: What Science, Mathematics, and Logic Cannot Tell Us” (The MIT Press) by Noson S. Yanofsky has been written that examines the issues of what is possible or not. I have found this book to be both informative and illuminative, and I would suggest that you read this book in conjunction with your readings and considerations on Artificial Intelligence. For those interested in the Brain versus the Mind question the book “Philosophy of Mind” 3rd Edition by Jaegwon Kim is extensive and comprehensive as this subject matter engenders. An interesting article on Artificial Intelligence is the “Benefits & Risks of Artificial Intelligence” from Futureoflife.org, that contains hyperlinks to other Artificial Intelligence writings of interest.
For many people who believe in the possibility of Artificial Intelligence, they often use “if” or “may” or “maybe” to justify their belief. But the use of “if” or “may” or “maybe” can be utilized to justify any belief. If and may and maybe can be utilized to justify the belief in Ancient Aliens affecting human civilizations, that UFO’s are Alien visitors, Paranormal activities happen, sightings of unknown creatures are factual occurrences, along with many other Pseudoscience and Superstitions. It is more important to determine if things are scientifically possible, and what are the constraints of the possible, to reach a justifiable belief.
Just because you hope that something may happen does not mean it will happen. Hope springs eternal, but reality will intervene. The hope of Artificial Intelligence is that it will provide answers to the questions that mankind has been searching for, as well as provide answers to questions that we did not even conceive. The reality is that the problems of Artificial Intelligence may be intractable or incomprehensible, and perhaps not even possible.
And remember, without the search for answers to the questions posed at the beginning of this article you do not have intelligence but merely an existence. Albeit an existence at a much higher level than animals.
Disclaimer
Please Note - many academics, scientist and engineers would critique what I have written here as not accurate nor through. I freely acknowledge that these critiques are correct. It was not my intentions to be accurate or through, as I am not qualified to give an accurate nor through description. My intention was to be understandable to a layperson so that they can grasp the concepts. Academics, scientists, and engineers entire education and training is based on accuracy and thoroughness, and as such, they strive for this accuracy and thoroughness. I believe it is essential for all laypersons to grasp the concepts of this paper, so they make more informed decisions on those areas of human endeavors that deal with this subject. As such, I did not strive for accuracy and thoroughness, only understandability.
Most academics, scientist, and engineers when speaking or writing for the general public (and many science writers as well) strive to be understandable to the general public. However, they often fall short on the understandability because of their commitment to accuracy and thoroughness, as well as some audience awareness factors. Their two biggest problems are accuracy and the audience knowledge of the topic.
Accuracy is a problem because academics, scientist, engineers and science writers are loath to be inaccurate. This is because they want the audience to obtain the correct information, and the possible negative repercussions amongst their colleagues and the scientific community at large if they are inaccurate. However, because modern science is complex this accuracy can, and often, leads to confusion amongst the audience.
The audience knowledge of the topic is important as most modern science is complex, with its own words, terminology, and basic concepts the audience is unfamiliar with, or they misinterpret. The audience becomes confused (even while smiling and lauding the academics, scientists, engineers or science writer), and the audience does not achieve understandability. Many times, the academics, scientists, engineers or science writer utilizes the scientific disciplines own words, terminology, and basic concepts without realizing the audience misinterpretations, or has no comprehension of these items.
It is for this reason that I place understandability as the highest priority in my writing, and I am willing to sacrifice accuracy and thoroughness to achieve understandability. There are many books, websites, and videos available that are more accurate and through. The subchapter on “Further Readings” also contains books on various subjects that can provide more accurate and thorough information. I leave it to the reader to decide if they want more accurate or through information and to seek out these books, websites, and videos for this information.
If you have any comments, concerns, critiques, or suggestions I can be reached at mwd@profitpages.com.
I will review reasoned and intellectual correspondence, and it is possible that I can change my mind,
or at least update the content of this article.